User-friendly activity recognition using SVM classifier and informative features
Phong C. H. Nguyen, Takayuki Akiyama, Hiroki Ohashi, Goh Nakahara, Katsuya Yamasaki, Saito Hikaru · 2015
For accurate indoor positioning, a moving activity recognition (AR) method has been developed that is based on a smartphone's sensor data. Prior methods can only recognize moving activities if the smartphone is held in a predefined place. We propose a method that works in various holding places to increase the usability. An SVM classifier is chosen because of its strength in utilizing features. New features are added such as percentiles of acceleration, air pressure, and acceleration magnitude. We have achieved 94.3% overall accuracy in various holding places: in users' hands, belt pouches, pant back pockets, and pant side pockets.